Munich Personal RePEc Archive
The J-Curve Dynamics of Turkey: An Application of ARDL Model
Halicioglu, Ferda
Department of Economics, Yeditepe University, Turkey
21 January 2008
Online at https://mpra.ub.uni-muenchen.de/6824/
MPRA Paper No. 6824, posted 15 Aug 2012 13:54 UTC
The J-Curve Dynamics of Turkey: An Application of ARDL Model
Abstract
This article seeks an empirical evidence for the existence of the J-curve phenomenon both in the short-run and long-run for Turkey over the period 1980-2005. The bounds testing cointegration approach is employed to estimate the trade balance model. An augmented form of Granger causality analysis is implemented between trade balance, real effective exchange rates, foreign income and domestic income. The stability of the short-run as well as long-run coefficients in the trade balance model is tested too. The empirical results suggest that the J-curve phenomenon is supported only in the short- run. Whilst causality tests reveal mix results, the parameter stability tests seem to be inconclusive.
Keywords: J-curve, trade balance, cointegration, causality, stability tests, Turkey JEL: C22, F14, F31
Correspondence to:
Ferda HALICIOGLU Department of Economics Yeditepe University Istanbul 34755 Turkey
fhalicioglu@yeditepe.edu.tr www.ferda-halicioglu.com
I. INTRODUCTION
A deficit in the trade balance of a country may be eliminated by a real devaluation in the domestic currency. Success of devaluation, however, depends on whether or not the sum of import and export elasticities exceeds unity, which is also known as the Marshall- Lerner (ML) condition. Bahmani-Oskooee (1985) argued that there have been cases under which the ML condition was satisfied yet the trade balance continued to deteriorate. Therefore, he concludes that the focus of a trade policy should be on the short-run dynamics that trace the post devaluation time path of the trade balance implying the J-curve phenomenon. The J-curve effect suggests that the trade balance of a nation initially worsens after a real depreciation of the home currency, and then it gets better. Krueger (1983) has argued that at the time an exchange occurs, goods, which are already in transit and under contract, have been purchased, and the completion of those transactions dominates the short-term change in the trade balance. Arndt and Dorrance (1987) indicates that this so called J-curve effect occurs if the domestic currency prices of exports are sticky. There has been growing interest in the J-curve phenomenon in the last three decades. Bahmani-Oskooee and Ratha (2004a) provides a very comprehensive survey on the J-curve literature for the period of 1973-2003 from thirty- seven articles. The recent examples of the J-curve studies include Bahmani-Oskooee et al. (2006), Bahmani-Oskooee et al. (2005), Bahmani-Oskooee and Ratha (2004b), Hacker and Hatemi (2004), Narayan (2004), and Narayan and Narayan (2004). The empirical literature on the evidence of the J-curve, however, is fairly ambiguous.
In regards to the empirical evidence for the Turkish J-curve, one can identify readily a few previous studies with inconclusive results. Rose (1990) finds that the real exchange rates have no impact on the trade balance. The work of Bahmani-Oskooee and Malixi (1992), which is based on Almon lag structure on real exchange rate, has found no support for it either. On employing the Engle-Granger cointegration approach, Bahmani-Oskooee and Alse (1994) asserts that the long-run impact of devaluation on the trade balance model is positive. Brada et al. (1997) who divided the data set into two sub-samples reports no long-run relationship between the variables of the trade balance function in the 1970s but they have revealed reverse results for the 1980s. Kale (2001) points out that a real depreciation of the domestic currency helps to improve the trade balance with a lag of about one-year and the impacts of devaluations on the trade balance are positive in the long-run. In a recent study, Akbostanci (2004) also presents empirical evidence of the J-curve phenomenon in the long-run. As a passing note, one should point out that the last two studies are based on that of Johansen-Juselius (1990) multivariate cointegration procedure with relatively short data spans derived from the 1980s and 1990s.
Turkey has adopted a policy of trade liberalization since 1980 and has pursued a successful export-led growth policy. The ratio of total exports to gross domestic product (GDP) increased from 4.1 to 13.3 percent during the period 1980-1988 and the real GDP grew by 5.8 percent in the same period. In 1989, there was a policy reversal, which slowed the depreciation of the Turkish lira (TL), in part to control inflation, but mainly to be able to easily borrow from the domestic markets, which led to five recessions in 1989, 1991, 1994, 1999, and 2001. The recessions of 1991, 1994 and 2001 were preceded by substantial increases in the real exchange rates. The impacts of devaluations on the trade balance, however, seem to be short lived as the early improvements in the trade balance are reversed steadily after a while. To combat the spiralling twin deficits in 2001, the IMF-led stabilization policy was put into effect once more. As a result, the internal imbalance has improved considerably but at the
same time the external balance has got worse. Despite having the free-floating exchange regime, the TL has been steadily appreciating against the major world currencies in real terms since 2002. This situation is being attributed to excess real domestic interest rates that are intentionally set at high levels to prevent inflation rising again. As a direct consequence of the overvalued TL, the current account deficit, which stems largely from the trade account deficit, has currently exceeded 6% of that Turkish GDP. Ertugrul and Selcuk (2001) provides a detailed account of the causes and consequences of the Turkish twin deficits in the 1980s and 1990s. A similar account of the Turkish economy beyond 2000 is given in Akyurek (2006).
The motivation of this study is two fold: persistent trade balance deficits of Turkey since the 1980s provide a good rationale to revisit the J-curve phenomenon and the cointegration procedure in this paper has been used for many countries in testing the J- curve phenomenon, except for Turkey.
The objectives of this study are as follows: i) to investigate the existence of the J-curve phenomenon both in the short-run and long-run using recent advances in time-series econometrics; ii) to establish the direction of causal relationships between trade balance, real effective exchange rates, foreign and domestic incomes variables; and iii) to implement parameter stability tests of Brown et al. (1975) to ascertain stability or instability in the trade balance model.
The remainder of this paper is organized as follows. Section II describes the study’s model and methodology. Section III discusses the empirical results, and finally Section IV concludes.
II. THE MODEL AND METHODOLOGY
The trade balance model employed in this study adopted the form of Rose and Yellen (1989) and it takes the following long-run (cointegrating) form:
t t t
t
t a a REER a YW a YT
TB = + ln + ln + ln +ε
ln 0 1 2 3 (1)
where the measure of the trade balance, TB is the ratio of imports to exports; REER is the real effective exchange rate; YW is the industrial production index of industrial countries; YT is the industrial production index of Turkey. Ln is the natural logarithm transformation and εt is the random error term. According to the J-curve phenomenon, it is expected that since an increase in real effective exchange rate initially reduces the demand for the home country’s export but increases its demand for imports.
As a result, the balance of trade worsens initially but it will improve after a while as export and import volumes adjust to price changes. While there are no apriori expectations about the signs of and , however, one asserts tentatively that is negative and is positive.
1<0 a
a2 a3 a2
a3
On investigating the cointegrating trade balance model with a view of testing the J- curve phenomenon, several econometric methods were implemented in the last two decades. In regards to univariate cointegration approaches, there are several examples including Engle-Granger (1987) and the fully modified OLS procedures of Phillips and Hansen’s (1990). In terms of multivariate cointegration, Johansen (1988), Johansen and Juselius (1990), and Johansen’s (1996) full information maximum likelihood procedures are widely employed. See Bahmani-Oskooee and Ratha (2004a) for a full account of the econometric procedures applied in the J-curve studies. A recent single cointegration approach, known as autoregressive-distributed lag (ARDL) of Pesaran
and Shin (1999) and Pesaran et al. (2001), has become popular amongst the researchers. Pesaran et al., cointegration approach, also named as bounds testing, has certain econometric advantages in comparison to other single cointegration procedures.
Firstly, endogeneity problems and inability to test hypotheses on the estimated coefficients in the long-run associated with the Engle-Granger method are avoided.
Secondly, the long and short-run parameters of the model in question are estimated simultaneously. Thirdly, all variables are assumed to be endogenous. Fourthly, the econometric methodology is relieved of the burden of establishing the order of integration amongst the variables and of pre-testing for unit roots. The ARDL approach to testing for the existence of a long-run relationship between the variables in levels is applicable irrespective of whether the underlying regressors are purely I(0), purely I(1), or fractionally integrated. Finally, according to Narayan (2004), the small sample properties of the bounds testing approach are far superior to that of multivariate cointegration.
An ARDL representation of equation (1) is formulated as follows:
t t t
t t
m
i
m
i
m
i
i t i m
i
i t i
i t i
i t i
t
v YT a YW a REER a
TB a
YT a
YW a
REER a
TB a
a TB
+ +
+ +
+
Δ +
Δ +
Δ +
Δ +
= Δ
−
−
−
−
= = = −
= −
−
∑
−∑ ∑ ∑
1 8 1 7
1 6
1 5
1 0 0
4 0
3 2
1 0
ln ln
ln ln
ln ln
ln ln
ln (2)
Pesaran et al. cointegration procedure is briefly outlined as follows. The bounds testing procedure is based on the F or Wald-statistics and is the first stage of the ARDL cointegration method. Accordingly, a joint significance test that implies no cointegration, (H0:a5 =a6 =a7 =a8 =0), should be performed for equation (2). The F test used for this procedure has a non-standard distribution. Thus, Pesaran et al.
compute two sets of critical values for a given significance level. One set assumes that all variables are I(0) and the other set assumes they are all I(1). If the computed F- statistic exceeds the upper critical bounds value, then the H0 is rejected. If the F-statistic falls into the bounds then the test becomes inconclusive. Lastly, if the F-statistic is below the lower critical bounds value, it implies no cointegration.
Once a long-run relationship has been established, equation (2) is estimated using an appropriate lag selection criterion. At the second stage of the ARDL cointegration procedure, it is also possible to perform a parameter stability test for the selected ARDL representation of the error correction model.
A general error correction model (ECM) of equation (2) is formulated as follows:
t t k
i
i t i
k
i
k
i
k
i
i t i
i t i
i t i
t
u EC YT
b
YW b
REER b
TB b
b TB
+ +
Δ +
Δ +
Δ +
Δ +
= Δ
= − −
= − = − = −
∑
∑ ∑ ∑
1 0
4
1 0 0
3 2
1 0
ln
ln ln
ln ln
λ
(3)
where λ is the speed of adjustment parameter and EC is the residuals that are obtained from the estimated cointegration model of equation (1).
The Granger representation theorem suggests that there will be Granger causality in at least one direction if there exists a cointegration relationship among the variables in equation (1), providing that they are integrated order of one. Engle-Granger (1987) cautions that the Granger causality test, which is conducted in first difference via a vector autoregression (VAR), will be misleading in the presence of cointegration.
Therefore, an inclusion of an additional variable to the VAR system, such as the error correction term would help us to capture the long-run relationship. To this end, an augmented form of Granger causality test involving the error correction term is formulated in a multivariate pth order vector error correction model.
[ ]
⎥⎥
⎥⎥
⎦
⎤
⎢⎢
⎢⎢
⎣
⎡ +
⎥⎥
⎥⎥
⎦
⎤
⎢⎢
⎢⎢
⎣
⎡ +
⎥⎥
⎥⎥
⎦
⎤
⎢⎢
⎢⎢
⎣
⎡
Δ Δ Δ Δ
⎥⎥
⎥⎥
⎦
⎤
⎢⎢
⎢⎢
⎣
⎡ +
⎥⎥
⎥⎥
⎦
⎤
⎢⎢
⎢⎢
⎣
⎡
=
⎥⎥
⎥⎥
⎦
⎤
⎢⎢
⎢⎢
⎣
⎡
Δ Δ Δ Δ
−
−
−
−
−
∑
=t t t t
t
i t
i t
i t i t
i i i i
i i i i
i i i i
i i i i p
i t
t t t
EC YT
YW REER TB
d d d d
d d d d
d d d d
d d d d
c c c c
YT YW
REER TB
4 3 2 1
1
4 3 2 1
44 43 42 41
34 33 32 31
24 23 22 21
14 13 12 11
1 4 3 2 1
ln ln ln ln
ln ln ln ln
ω ω ω ω
λ λ λ λ
(4)
ECt-1 is the error correction term, which is derived from the long-run relationship, and it is not included in equation (4) if one finds no cointegration amongst the variables in question. The Granger causality test may be applied to equation (4) as follows: i) by checking statistical significance of the lagged differences of the variables for each vector; this is a measure of short-run causality; and ii) by examining statistical significance of the error correction term for the vector that there exists a long-run relationship.
The existence of a cointegration derived from equation (2) does not necessarily imply that the estimated coefficients are stable as argued in Bahmani-Oskooee and Brooks (1999). Hence, stability tests of Brown et al. (1975), which are also known as cumulative sum (CUSUM) and cumulative sum of squares (CUSUMSQ) tests based on the recursive regression residuals, may be employed to that end. These tests also incorporate the short-run dynamics to the long-run through residuals.The CUSUM and CUSUMSQ statistics are updated recursively and plotted against the break points of the model. Providing that the plot of these statistics fall inside the critical bounds of 5%
significance, one assumes that the coefficients of a given regression are stable. These tests are usually implemented by means of graphical representation.
III. THE EMPIRICAL RESULTS
Quarterly data over 1980I-2005IV period were used to estimate equation (2). Data definition and sources of data are cited in the appendix. Equation (2) was estimated in two stages. In the first stage of the ARDL procedure, the order of lags on the first–
differenced variables for equation (2) was obtained from unrestricted VAR by means of Akaike Information Criterion (AIC) and Schwarz Bayesian Criterion (SBC), which indicated the optimal lag level as four and six quarters respectively for this study. The results of this stage are not reported here for brevity. Then an F deletion test was applied to equation (2) in order to test the existence of a long-run relationship by using lags from four to six on following Bahmani-Oskooee and Goswami (2003). As they have shown that the results of this stage are sensitive to the order of VAR. Equation (2) was also estimated three more times in the same way but the dependent variable each time was replaced by one of the explanatory variables in search of other possible long- run relationship in any other form than it had already been described in equation (1).
The summary results of bounds tests are presented in Table 1. Table 1 indicates only one plausible a long-run relationship in which lnTB is dependent variable.
Table1. F-statistics for cointegration relationship
Critical value bounds of the F-statistic 90% level 95% level I(0) I(1) I(0) I(1) 2.721 3.773 3.232 4.351
Calculated F-statistics
F(lnTB│lnREER, lnYW, lnYT) =5.175*
F(lnREER│lnTB, lnYW, lnYT) =1.412 F(lnYW│lnREER, lnTB, lnYT) =3.283 F(lnYT│lnREER, lnYW, lnTB) =3.104
The relevant critical values are obtained from Table C1.iii (with an unrestricted intercept and no trend with three regressors) in Pesaran et al. (2001).* indicates the statistical significance at the 5 % level. The optimal lag length is four.
In the second stage of ARDL cointegration procedure, equations (2) and (3) were estimated respectively on the basis of AIC and SBC criteria with six lags in search of selecting best-fitted model. The results of the model selection stages are reported in Table 2 and 3 respectively.
Table 2. Long-run coefficients of the trade balance of Turkey Panel A: the long-run results
Dependent variable lnTB
Model Selection Criterion
Regressors AIC ARDL (1,2,3,0) SBC ARDL (1,1,3,0) REER
ln 0.468 (1.449) 0.656* (2.068)
lnYW -3.123* (2.326) -3.440* (2.477) YT
ln 1.328* (2.371) 1.491* (2.572) Constant 7.387 (2.081) 7.186 (1.989) Panel B: the short-run diagnostic test statistics
2
χSC(4)=2.324, χ2FC(1)=5.731 χSC2 (4)=2.144, χFC2 (1)=4.948
2
χN(2)=7.192, χH2 (1)=0.325
2
χN(2)=8.176, χH2 (1)=0.206
Notes: The absolute value of t-ratios is in parentheses. , , , and are Lagrange multiplier statistics for tests of residual correlation, functional form mis-specification, non-normal errors and heteroskedasticity, respectively. These statistics are distributed as Chi-squared variates with degrees of freedom in parentheses. * indicates statistical significance at the 5% level.
2
χSC χ2FC χ2N χH2
Considering the long-run coefficients in the panel A of Table 2, one concludes that the SBC based trade balance model seems to be a slightly better fit than the AIC model since all the coefficients in the former model are statistically significant at the 5% level.
Apart from that difference, both model selection criteria reveal almost identical results.
Table 3. Short-run coefficients of the trade balance of Turkey- estimates from ECM
Dependent variable ΔlnTBt
Model Selection Criterion Regressors AIC ARDL (1,2,3,0) SBC ARDL (1,1,3,0)
REERt
Δln 0.907* (3.759) 0.952* (3.890)
ln −1
Δ REERt 0.494* (2.044) - -
YWt
Δln -2.323* (4.011) -2.225* (3.787)
ln −1
Δ YWt 0.831 (1.276) 1.116 (1.723)
ln −2
Δ YWt 2.845* (5.138) 2.939* (5.233)
YTt
Δln 0.467* (2.525) 0.522* (2.804)
Constant 2.600 (2.251) 2.517 (2.143)
−1
ECt -0.351* (4.938) -0.350* (4.830)
R2 0.481 0.507
F-statistics 14.168 15.287
DW-statistics 2.034 2.041
RSS 2.071 2.131
Notes: The absolute values of t-ratios are in parentheses. RSS stands for residual sum of squares.
* indicates statistical significance at the 5% level.
As far as the ECMs are concerned, it can be seen from Table 3 that the results of equation (3) are also very close in both model selection criteria. The long-run and short run results of the SBC version of the trade balance model, however, provide again relatively better results in terms of the expected signs and econometric diagnostics. The speed of adjustment coefficient, -0.35 is considerably low indicating a slow convergence to equilibrium in the case of a shock to the cointegrating relationship and that long-run equilibrium is attained only after three quarters. Based on the AIC and SBC results, however, one can suggest that there are no J-curve effects both in the long and short-runs since the coefficient of the real effective exchange rates does not alter from positive to negative in either case. Considering that the focus of this paper is essentially on the dynamics of devaluation, the impact of the lags of the real exchange rates on the trade balance from error correction version of the SBC model was estimated. The results are reported in Table 4. The evidence of the J-curve is now apparent in the REER variable as the positive coefficients are followed by the negative coefficients. This pattern of the coefficients shows an initial deterioration followed by an improvement in the trade balance.
Table 4. Coefficient estimates of ΔlnREERt−i and error correction term Dependent variable ΔlnTBt
Regressors Coefficients t-statistics REERt
Δln 0.890* 3.144
ln −1
Δ REERt 0.370 1.308
ln −2
Δ REERt 0.323 1.115
ln −3
Δ REERt 0.327 1.141
ln −4
Δ REERt -0.094 0.319
ln −5
Δ REERt -0.240 0.838
ln −6
Δ REERt 0.403 1.407
Constant -0.018 0.961
−1
ECt -0.327* 4.102
R2 0.421
Diagnostic test statistics
2
χSC(4)=5.129 χFC2 (1)=2.623,χN2(2)=4.143, χH2 (1)=0.003
Notes: t-ratios are in absolute values. , , , and are Lagrange multiplier statistics for tests of residual correlation, functional form mis-specification, non-normal errors and heteroskedasticity, respectively. These statistics are distributed as Chi-squared variates with degrees of freedom in parentheses. * indicates statistical significance at the 5% level.
2
χSC χFC2 χN2 χH2
Granger causality test requires that all variables are in equation (1) should be in integrated of order one. To this end, the traditional unit root tests such as the augmented Dickey and Fuller (1979, 1981) and the Phillips and Peron (1988) were employed. The unit root test results suggest that all variables in their first level differences are I (1). For brevity of presentation, they are not reported here. Having a cointegrating relationship among [TBt, REERt, YWt, YTt] on the basis of the results of the bounds test in Table 1, Granger causality test was conducted to equation (4) as such that only the trade balance vector was estimated with an error correction term. Similarly the Granger causality tests were applied to other models without the error correction terms since one could not ascertain any long-run relationship for the other vectors. Table 5 summarises the results of long-run and short-run Granger causality.
Table 5. Results of Granger Causality
F statistics (probability)
Dependent Variable ΔlnTBt ΔlnREERt ΔlnYWt ΔlnYTt ECt−1 (t-statistics) TBt
Δln - 2.94*
(0.01)
4.82*
(0.01)
1.54 (0.18)
-0.26*
(3.24) REERt
Δln 3.32*
(0.01)
- 1.42 (0.22)
1.46 (0.20)
- YWt
Δln 0.79
(0.55)
0.65 (0.66)
- 2.18 (0.06)
- YTt
Δln 1.64
(0.15)
3.40*
(0.00)
2.97*
(0.01)
- - Causality inference: REER⇔TB, YW⇒TB, YW⇒YT, REER⇒YT
Notes:* indicates statistical significance at the 5% level. The optimal length is 4 and is based on SBC.
On considering the long-run Granger causality, one suggests that the lagged error correction term with the expected sign is statistically significant which also confirms the results of the bounds test. The speed of the adjustment, however, is less than the ARLD estimations suggesting that the long-term convergence to equilibrium will take longer. Granger causality runs interactively via both exchange rates and world income to trade balance ratio. In regards to the short-run Granger causality tests based on the F statistics, one sees that there exists a bi-directional Granger causality between exchange rates and trade balance ratio, implying that as the trade balance improves it causes a real appreciation of the TL. Likewise, a real depreciation will have a positive impact on the trade balance. There are, however, uni-directional Granger causalities, which run from world income to the trade balance, from world income to domestic income, and from exchange rates to domestic income.
Plot of Cumulative Sum of Recursive Residuals
The straight lines represent critical bounds at 5% significance level
-5 -10 -15 -20 -25 0 5 10 15 20 25
1981Q2 1985Q1 1988Q4 1992Q3 1996Q2 2000Q1 2003Q4 2005Q4
Figure 1. Plot of CUSUM
Plot of Cumulative Sum of Squares of Recursive Residuals
The straight lines represent critical bounds at 5% significance level -0.5
0.0 0.5 1.0 1.5
1981Q2 1985Q1 1988Q4 1992Q3 1996Q2 2000Q1 2003Q4 2005Q4
Figure 2. Plot of CUSUMSQ
The parameter stability tests were implemented via equation (3), which captures the short-run dynamics of equation (2) and the long-run impact of equation (1). Equation (3) was estimated by ordinary least squares with a lag length of four based on the SBC criterion. Figure 1 and 2 provides the plots of CUSUM and CUSUMSQ tests, respectively. Figure 1 and 2 display conflicting results in regards to the parameter stability tests. Whilst the former test indicates instability the later suggests stability in the parameters of the trade balance equation.
IV. CONCLUSION
This study has attempted to estimate the J-curve phenomenon through a reduced form trade balance model in search of providing fresh empirical evidence in the case of Turkish data. A recent single cointegration technique proposed by Pesaran et al. (2001) was employed to investigate the short-run and long-run responses of the trade balance to currency depreciation. The results from bounds tests have indicated only one possible long-run relationship in which the trade balance is the dependent variable. In contrast to the results of the previous results relating to Turkey, one finds no long-run impact of a real devaluation on the trade balance but there exists the J-curve phenomenon in the short-run. The existence of J-curve in the short-run confirms that the Turkish foreign trade balance deficit improves rather rapidly on following currency depreciations as it has happened in the TL depreciations of 1994, 1999 and 2004.Whilst, in the long-run, real effective exchange rates and world income Granger cause to the trade balance, in the short-run there is a feedback relationship between the real effective exchange rates and trade balance. The parameter stability tests on the long-run trade balance equation appear to be inconclusive.
APPENDIX
Data definition and sources
The data set used in this study cover the period 1980I to 2005IV. All data are collected from International Financial Statistics, (IMF) and Central Bank of Turkey (CBT).
TB is Turkey’s trade balance. It is defined as the ratio of real imports to real exports.
The raw data of exports and imports were deflated by the USA consumer price index.
Source: IMF.
REER is the real effective exchange rates index. Source: CBT.
YW is industrial production index of industrial countries. Source: IMF.
YT is industrial production index of Turkey. Source: IMF
REFERENCES
Akbostanci, E. (2004) Dynamics of the trade balance: the Turkish J-curve, Emerging Markets Finance and Trade, 40, 57-73.
Akyurek, C. (2006) The Turkish crisis of 2001: a classic?, Emerging Markets Finance and Trade, 42, 5-32.
Arndt, H. W. and Dorrance, G. (1987) The J-curve, The Australian Economic Review, 11, 9-19.
Bahmani-Oskooee, M. (1985) Devaluation and the J-curve: some evidence from LDCs, The Review of Economics and Statistics, 17, 500-504.
Bahmani-Oskooee, M. and Alse, J. (1994) Short-run versus long-run effects of devaluation: error correction modelling and cointegration, Eastern Economic Journal, 20, 453-464.
Bahmani-Oskooee, M. and Brooks, T. J. (1999) Bilateral J-curve between US and her trading partners, Weltwirtschaftliches Archiv, 135, 156-165.
Bahmani-Oskooee, M. and Malixi, M. (1992) More evidence on the J-curve from LDCs, Journal of Policy Modeling, 14, 641-653.
Bahmani-Oskooee, M. and Goswami, G. G. (2003) A disaggregated approach to test the J-curve phenomenon: Japan versus her major trading partners, International Journal of Economics and Finance, 27, 102-113.
Bahmani-Oskooee, M. and Ratha, A. (2004a) The J-curve: a literature review, Applied Economics, 36, 1377-1398.
Bahmani-Oskooee, M. and Ratha, A. (2004b) The J-curve dynamics of U.S. bilateral trade, Journal of Economics and Finance, 28, 32-38.
Bahmani-Oskooee, M., Goswami, G.G., and Talukdar, B. K. (2005) The Bilateral J- curve: Australia versus her 23 trading partners, Australian Economic Papers, 44, 110- 120.
Bahmani-Oskooee, M. C. Economidou, G.G. Goswami (2006) Bilateral J-curve between the UK vis-à-vis her major trading partners, Applied Economics, 38, 879-888.
Brada, J.C., Kutan, A.H., and Zhou, S. (1997) The exchange rate and the balance of trade: the Turkish experience, Journal of Development Studies, 33, 675-692.
Brown, R.L., Durbin, J. and Evans, J.M. (1975) Techniques for testing the constancy of regression relations over time, Journal of the Royal Statistical Society, 37, 149-163.
Dickey, D.A., and Fuller, W.A., (1979) Distributions of the estimators for autoregressive time series with a unit root, Journal of the American Statistical Association, 74, 427-431.
Dickey, D.A., and Fuller, W.A. (1981) Likelihood ratio statistics for autoregressive time series with a unit root, Econometrica, 49, 1057-1072
Ertugrul, A. and Selcuk, F. (2001) A brief account of the Turkish economy, 1980-2000, Emerging Markets Finance and Trade, 37, 6-30.
Engle, R. and Granger, C. (1987) Cointegration and error correction representation:
estimation and testing, Econometrica, 55, 251-276.
Hacker, R.S. and Hatemi, A.J. (2004) The effect of exchange rate changes on trade balances in the short and long run, Economics of Transition, 12, 777-799.
Johansen, S. (1988) Statistical analysis of cointegrating vectors, Journal of Economic Dynamics and Control, .12, 231-254.
Johansen, S. (1996) Likelihood-Based Inference in Cointegrated Vector Auto- Regressive Models, 2nd ed., Oxford University Press, Oxford.
Johansen, S. and Juselius, K. (1990) Maximum likelihood estimation and inference on cointegration–with application to the demand for money, Oxford Bulletin of Economics and Statistics, 52, 169-210.
Kale, P. (2001) Turkey’s trade balance in the short and the long run: an error correction modelling and cointegration, The International Trade Journal, 15, 27-56.
Krueger, A.D. (1983) Exchange Rate Determination Cambridge University Press, Cambridge.
Narayan, P.K. (2004) New Zealand’s trade balance: evidence from the J-curve and granger causality, Applied Economics Letters, 11, 351-354.
Narayan, P.K. and Narayan, S. (2004) The J-curve: evidence from Fiji, International Journal of Applied Economics, 18, 369-380.
Peseran, M. H. and Shin, Y. (1999) An autoregressive distributed lag modelling approach to cointegration analysis, In Strom, S. (Ed.): Econometrics and Economic Theory in 20th Century: The Ragnar Frisch Centennial Symposium, Chapter 11, Cambridge University Press, Cambridge.
Pesaran, M.H., Shin, Y., and Smith, R.J. (2001) Bounds testing approaches to the analysis of level relationships, Journal of Applied Econometrics, 16, 289-326.
Phillips, P.C. B. and Perron, P. (1988) Testing for a unit root in time series regression, Biometrika, 75, 335-359.
Phillips, P. and Hansen, B. (1990) Statistical inference in instrumental variables regression with I(1) process, Review of Economic Studies, 57, 99-125.
Rose, A. K. (1990) Exchange rates and the trade balance: some evidence from developing countries, Economics Letters, 34, 271-275.
Rose, A. K. and Yellen, J. L. (1989) Is there a J-curve?, Journal of Monetary Economics, 24, 53-58.
